Measuring
Measuring marketing without analytics: four numbers, one sheet
Some sections describe the law in Germany (as of September 2026). General information, not legal advice. Other countries regulate this differently.
You do not need an analytics tool to know which channel works, because the number that decides it is not visits but paying customers, and no tool can see why someone bought. Four numbers are enough: where each paying customer came from, what the channel cost in money and in your own hours, how many customers it produced, and what one customer contributes after their own costs. Divide the full cost by the customers and you have the cost per customer; hold that against the contribution margin and the channel has answered for itself. The source comes from one optional open question at the point of purchase, which needs no consent banner because nothing is stored on or read from anyone's device. Ten minutes a month in a spreadsheet beats a dashboard that only ever saw the visitors who clicked accept.
The tool measures the wrong thing, and only part of it
The usual first move is to install an analytics tool, put a consent banner on the site and wait for clarity. What arrives instead is a dashboard full of visits, sessions, bounce rates and durations, and not one of those numbers answers the question you actually have, which is: which channel brought me customers who paid, and what did they cost?
There is a second problem on top of the first. Analytics needs consent, and consent is given by some visitors and refused by others. The ones who refuse are not a random sample, so the tool does not show you a smaller version of reality, it shows you a tilted one. You end up with precise decimals describing a group you did not choose.
And even a perfect tool would see only the last click before the purchase. It cannot see the colleague who recommended you three weeks ago, the card someone kept, or the conversation at the counter. Those are exactly the channels that work early on. The customer knows all of it, and asking costs nothing.
Which channel suits which audience, with lead times and limits: marketing channels for founders →
Four numbers, and where each one comes from
Everything in this article runs on these four. Three of them you already have somewhere; only the first has to be collected, and it takes one sentence to collect. Keep them per channel and per month, because a channel only becomes comparable once it has a period attached to it.
What you write down, and where you get it
Where the customer came from
One word per paying customer: referral, search, Instagram, trade fair, neighbourhood. Not per visitor, not per enquiry, but per customer who paid. It is the only one of the four you cannot calculate, and the only one an analytics tool would not reliably give you either.
From the question in the first conversation or on the order form. It is further down.
What the channel cost
Two items, not one: the money you spent, and your own hours multiplied by what one of your hours is worth. The second is missing from almost every calculation founders make, and in almost every channel it is the larger one. A channel that "costs nothing" and takes forty hours a month is the most expensive one you have.
From your bank statement and a tally of hours. Neither needs a tool.
How many customers it produced
The count from row 1, added up per channel. Careful with small numbers: at three customers a month, one coincidence moves the result by a third. Below ten customers per channel you are mostly judging chance rather than the channel.
From your own sheet, added up once a month.
What a customer brings in
The contribution margin: what one customer pays, minus what that one customer costs you in materials, time and fees. Not revenue. Measure against revenue and every channel looks viable until the bank account says otherwise.
From your pricing. If you do not have one, that is the first step, not measuring.
Visits are a number about your website. Cost per customer is a number about your business.
Contribution margin means revenue from one customer minus the costs that customer alone causes. It is not profit, because it carries none of your fixed costs; it is the amount that is left to cover them.
The one question, and where it belongs
One optional open field, placed at the moment of purchase rather than before it: "How did you hear about us?" Not mandatory, not a dropdown, and not on the enquiry form, because at that point the person has not yet decided anything and the answer describes a visit rather than a sale.
Why open text and not a dropdown
A dropdown offers your guesses back to you. People pick the first plausible entry, the top option grows, and channels you never thought of stay invisible because there is no box for them. An open field costs the customer four seconds and gives you the wording.
The wording is the part that pays. "A colleague recommended you" and "I found you on the map" lead to entirely different next steps, and a dropdown would have compressed both into one line. You sort the answers into channels yourself, once a month, which takes minutes and keeps the original sentence intact.
A freely given optional answer that you store as a word, without linking it to the person, is not a consent question about terminal equipment at all: nothing is stored on the device and nothing is read from it. General orientation, not legal advice.
What one customer from a channel costs
Take one channel and one month. Put in the money, the hours, what an hour of yours is worth, the customers it produced and what one of them contributes. The tool then does the only calculation that matters here, and it does it twice: once with your time counted and once without, because the gap between those two is where most channel decisions go wrong.
What one customer from this channel costs
One channel, one month, five numbers from your own sheet. The starting values are examples, not benchmarks. Nothing is stored or transmitted.
Money spent per month
200 €
Your own hours per month
16 h
What one of your hours is worth
40 €
Paying customers from it per month
6
Contribution margin per customer
400 €
Full cost per customer
140 €
Counting money alone, you would put this at 33 €. The difference is your own time, and it is real.
What the channel actually costs
Money
200 € (24 %)
Your time
640 € (76 %)
Each customer brings in 400 € and costs 140 €, a ratio of 2.9 to 1. Below one to one you lose money on every single customer. That part is arithmetic. The widely quoted three to one is a rule of thumb, not a measured value: it exists so that there is room left for everything the channel does not pay for.
At this level of spend the channel needs 3 customers a month to pay for itself. Below that it is an investment you are making, which is fine as long as you know you are making it.
Your next step
This channel pays for itself 2.9 times over. The question is no longer whether it works but where it stops: put in ten percent more next month and read the cost per customer again. As long as it stays flat, keep going. The month it starts to climb is the month you have found the ceiling of this channel, and that is when a second one becomes worth starting.
The model is deliberately plain: full cost is money plus hours times your hourly value, cost per customer is full cost divided by customers, and the ratio is contribution margin divided by cost per customer. No industry figures are stored anywhere in this tool. The one arithmetic fact in it is that a ratio below one to one loses money on every customer; the three-to-one threshold in the colours is a widely used rule of thumb and not a measurement.
What works without a consent banner, and what does not
The dividing line is not "cookie or no cookie", which is where most founders place it. It runs between methods that touch the visitor's device and methods that do not. Everything in the first group needs consent; everything in the second does not, although personal data still has to be handled properly.
Does it need consent?
Asking, in conversation or on the form
no consent
Nothing is stored on the device and nothing is read from it, so the consent rule for terminal equipment does not apply at all. An optional field also needs no consent under data protection law as long as you keep the answer as a word and do not tie it to the person.
Reading your own server logs
no consent
Your server already records which page was requested and when. Counting those lines does not touch the visitor's device. The IP address is still personal data, so shorten it or do not keep it at all, and rest the analysis on legitimate interest.
Using a separate address per channel
no consent
A separate phone number, email address or landing page per channel separates the sources without tracking anything. The card in the window carries a different number from the advert. Whoever calls has sorted themselves.
A voucher code per channel
no consent
One code per channel attributes the sale without touching the device. It also tells you something no tool can measure: the customer did not merely see the channel, they remembered something from it.
An analytics tool with a cookie
consent
Consent required, with no exception. The statistics exception that once appeared in a European draft never became law. Loading the tool before consent breaches the rule.
Tracking pixels and tagged links
consent
The common assumption that "no cookie" means "no consent needed" is wrong. European regulators held in October 2024 that even caching the requested resource counts as storage on the device, which puts tracking pixels and tagged links under the same rule.
Device fingerprinting
consent
Reading screen resolution, fonts, time zone and extensions is access to the device and needs consent. That nothing is written changes nothing: the rule covers storing and accessing alike.
Consent through a recognised service
in progress
Since 1 April 2025 Germany has had a regulation for recognised consent management services: the user decides once, centrally, and individual sites stop asking. The public register lists exactly one recognised service so far, since October 2025. For a founder in 2026 that changes nothing yet.
Germany: § 25 Abs. 1 TDDDG requires consent for storing information on, or accessing information in, terminal equipment; the two exceptions in Abs. 2 cover transmission and what is strictly necessary for a service the user explicitly requested, neither of which covers audience measurement. Fines run to 300,000 euros under § 28 TDDDG. The European Data Protection Board adopted the final version of its guidelines on the technical scope of Art. 5(3) ePrivacy Directive on 16 October 2024, bringing tracking pixels and tagged links into scope. The German consent management regulation has been in force since 1 April 2025, with one recognised service listed since 17 October 2025. General orientation, not legal advice.
The sheet: six columns, ten minutes a month
One row per paying customer, filled in the week they pay rather than at the end of the month, because by then nobody remembers the wording. Once a month you sort the third column into the fourth, add up per channel and put the result into the calculator above. That is the whole system.
Date
14 Sep
Customer
M. Berger
Source (their words)
"A colleague recommended you"
Channel
Referral
Revenue
890 €
Contribution margin
410 €
Column three keeps the customer's own words, column four your classification of them. Keeping both is what lets you reclassify later without losing the original, and the original is where the reasons live.
Where the contribution margin in column six comes from: startup pricing in five steps →
Five mistakes that cost the most
The first two make the numbers wrong. The last three make correct numbers useless, which is the more expensive failure, because it looks like diligence.
Measuring visits instead of customers. Ten thousand visitors and two customers is a bad month, not good news. The number at the top of every tool is the one that says least about your business.
Leaving your own time out. It is the larger cost in almost every channel and the only one that never appears on a bank statement. That is precisely why the "free" channel is so often mistaken for the cheapest.
Judging too early. At three customers a month, one more or less is a third of the result. Change course after four weeks and you are reacting to chance, never learning what worked.
Changing everything at once. New advert, new page, new price, all in the same month. Afterwards the number is better or worse and nobody knows which change did it.
Measuring a slow channel by a fast one's clock. Search and referrals take months, an advert works in days. Comparing them over the same window regularly kills the channel that would have become the cheapest.
Frequently asked questions
Do I really not need an analytics tool?
Not for the question of which channel produces paying customers. The customer answers that one better than any tool: a tool sees the last click before the purchase, the customer tells you the reason. Tools are useful for other questions, such as where people drop out of a long checkout. That is rarely the question that decides survival in the first months.
How do I ask without being pushy?
One optional open field at the very end of checkout, or a last sentence in the first conversation: "How did you hear about us?" Not mandatory, not a dropdown. Dropdowns push people towards whatever sits at the top, and that option then looks larger than it is. The wording of the answer is worth more than the category, because it contains what convinced them.
What about customers who saw several channels?
That is the normal case, and no method resolves it cleanly, including expensive ones. The pragmatic answer: record what the customer names first and accept that attribution is imprecise. An imprecise number you have beats a precise one you do not. For larger purchases you can also ask what tipped the decision.
When does the number become reliable?
At roughly ten customers per channel an anecdote becomes a tendency. Below that, a single coincidence moves the result so much that the number says more about the month than about the channel. That is not a reason to skip measuring, it is a reason not to change course after four weeks.
Is a cookieless tracking pixel not enough?
No, and this is the most common misconception in the area. European regulators held in October 2024 that caching a requested resource already counts as storage on the device, which puts tracking pixels and tagged links under the same consent requirement as cookies. General orientation, not legal advice.
What is a good cost per customer?
There is no universal figure. The only benchmark is your own contribution margin: if a customer costs more than they bring in, you lose money on every one, and the better the channel runs the more you lose. The often quoted three-to-one ratio is a rule of thumb, not a measurement. It exists so that something is left over for rent, tools and you.
Read next
Is the idea worth the marketing?
No measurement saves a channel that is selling the wrong thing. 4 to 12 AI experts assess market, financials and risks of your specific idea, in minutes, from € 5.99.
Assess my idea →